DeepSeek Harness Plugins · Workflow
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
A file-browser web app for DeepSeek Harness with multi-view navigation, search, and one-click AI analysis.
dsh-foundry is a self-bootstrapping plugin compiler for DSH, simplifying plugin development with a blueprint system for one-click compilation and static validation.
dsh-agent-arena is a multi-AI collaboration plugin that puts humans and multiple AIs in a single persistent chat session for parallel task processing, tool usage, and management via task boards, decision boards, and outcome repositories.
dsh-fabric provides a declarative DSL and tools for DeepSeek Harness to unify runtime extensions, enabling dynamic plugin registration and querying.
A Windows-adapted coding mode agent preset for DeepSeek Harness, providing a persistent PowerShell 7 shell and editor tool.
A multi-agent orchestration runtime plugin that defines agent teams with six collaboration strategies and structured task graphs, usable via Python API, CLI, HTTP, or MCP stdio.
A DSH plugin for discussion mode that maintains user intent, controls focus via a four-line Rail, and converges exploration into evidence-based next steps.
Integrates Kimi Code Swarm mode into DeepSeek Harness for batch parallel subtask dispatch with real-time progress tracking, enhancing efficiency.
A green interrupt button that silently halts the agent, which then summarizes its progress and asks for new requirements, with customizable interrupt prompts.
A DSH plugin for AI role-playing and multi-role discussions using @ mentions, where each role responds independently and a moderator converges consensus, disagreements, and action suggestions.
catalog-capabilities-zh is a public capability directory for Codex and DeepSeek Harness that maintains Chinese documentation and orchestrates trusted installers.
dsh-capability-index is a meta-plugin for DeepSeek Harness that pre-checks the plugin library and injects applicable tool hints via semantic and rule-based methods to improve utilization, with final decisions by the model.